> The structure of the LEAF is also a dead giveaway that Clipper is > being used -- it's easy to envision a box that has the family key, > and tries every LEAF-sized field to see if it decrypts to something > that looks right, and in particular has the right checksum. I'm going to make the almost certainly valid assumption that you know more about the way the network works than I do, but my assumption is this: in the wacky scenario I described wherein Clipper devices are installed in the network interfaces "everywhere", then the presence of these identifiable (and identifying!) packets means that a central tap at a regional switching center could concievably perform traffic analysis without the need for taps on local loops anywhere. Is this assumption way wrong? I suspect that you'd have too much data -- you'd have to be able to scan every part of every conversation. If you're going to go to those lengths, you'd do just as well to tap the signaling channels instead -- a lot less data, and most of it organized the way you want it.